RESEARCH PAPER NO. 14 - Economic Complexity and Growth: Can value-added exports better explain the link? - EcoAustria
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
November 2020
RESEARCH PAPER NO. 14
Economic Complexity and Growth: Can value-added
exports better explain the link?
Philipp Koch
www.ecoaustria.ac.atRESEARCH PAPER NO. 14 Economic Complexity and Growth: Can value-added exports better explain the link? Philipp Koch, MSc. – EcoAustria - Institute for Economic Research November 2020 Imprint: EcoAustria – Institute for Economic Research, Am Heumarkt 10, 1030 Vienna, Austria, Tel: +43-(0)1-388 55 11 www.ecoaustria.ac.at
Economic Complexity and Growth: Can value-added exports better
explain the link?
Philipp Koch1
November 2020
Abstract
In economic literature, economic complexity is typically approximated on the basis of an economy’s gross
export structure. However, in times of ever increasingly integrated global value chains, gross exports may
convey an inaccurate image of a country’s economic performance since they also incorporate foreign value-
added and double-counted exports. Thus, I introduce a new empirical approach approximating economic
complexity based on a country’s value-added export structure. This approach leads to substantially different
complexity rankings compared to established metrics. Moreover, the explanatory power of GDP per capita
growth rates for a sample of 40 lower-middle to high-income countries is considerably higher, even if
controlling for typical growth regression covariates.
Keywords: complexity, economic growth, value-added exports
JEL: O19, O47, F43
1
EcoAustria – Institute for Economic Research, Vienna (Austria);
Email: philipp.koch@ecoaustria.ac.at1 Introduction
Economic complexity and its relationship to economic growth is an increasingly re-
searched topic, originating in seminal contributions by Hidalgo, Klinger, Barabási, and
Hausmann (2007), Hausmann, Hwang, and Rodrik (2007), and Hidalgo and Hausmann
(2009), who show that economic complexity can explain cross-country income differences
and predict future growth rates.
A country is said to be complex if (i) it is diversified, i.e. is able to produce and
export a wide range of products; and (ii) if it exports less ubiquitous products, which
are assumed to be more complex than ubiquitous products. By taking both dimensions
into account, economic complexity can be seen as a latent measure of the amount of
productive knowledge a country holds, which influences economic performance. Hence,
complexity is intertwined with, but goes beyond the concept of human capital.
Existing studies approximate an economy’s complexity by its gross export structure
in terms of products. However, as Koopman, Wang, and Wei (2014) point out, gross
exports do not only capture the value that is added in the exporting country, but also the
foreign value-added that was imported as intermediate goods, exported goods that are
eventually consumed in the domestic country, as well as double-counted exports. Since
the share of foreign value added in exports increased over the past decades (Johnson
& Noguera, 2017), gross exports may lead to incorrect conclusions when analyzing a
country’s economic performance (e.g. Timmer, Miroudot, & de Vries, 2019).
To take this shortcoming into account, this article contributes to the literature by
proposing to approximate an economy’s complexity based on the structure of its value-
added exports, which are defined as domestic value added in exports of intermediary or
final goods that are eventually consumed in a foreign country. The performance of the
complexity measure based on value-added exports in explaining GDP per capita growth
is then compared to two established indices - the Economic Complexity Index (ECI )
by Hidalgo and Hausmann (2009) and Economic Fitness (EF ) proposed in Tacchella,
Cristelli, Caldarelli, Gabrielli, and Pietronero (2012) and Tacchella, Cristelli, Caldarelli,
Gabrielli, and Pietronero (2013).
2 Data & Methodology
Data. Value-added exports are calculated based on the most recent release of the
World-Input-Output Database (Timmer, Dietzenbacher, Los, Stehrer, & de Vries, 2015),
2covering 56 industries in 43 lower-middle- to high-income annually from 2000 to 2014.1,2
For data on GDP per capita, capital, and population I rely on the Penn World Table
(Feenstra, Inklaar, & Timmer, 2015). Data on human capital are obtained from the
Wittgenstein Centre for Demography and Global Human Capital (2018). Additionally,
data on the ECI are provided by The Growth Lab at Harvard University (2019), while
data on EF can be obtained from the World Bank data catalog.
Methodology. The ECI and EF are calculated based on a binary adjacency matrix,
which denotes whether a country has a Revealed Comparative Advantage (Balassa, 1965)
in a specific product in terms of gross exports.3 However, since value-added exports
only refer to industries and, thus, the dimensions of the adjacency matrix are reduced
considerably, a binary adjacency matrix does not introduce satisfactory variation over
time or within cross-sections, since specializations in value-added exports rarely change.
Hence, I apply a weighted adjacency matrix W.4 Its elements are defined as the share
of value-added exports (V X) country c has in industry s, i.e.
V Xcs
Wcs = P (1)
c V Xcs
This adjacency matrix allows for calculating EF in terms of value-added exports, in the
following referred to as value-added export Fitness (V XF ).5 Analogous to Tacchella et
al. (2012) and Tacchella et al. (2013), it is defined as an iterative process of order N
such that
F̃c,N
V XFc,N = 1 P (2)
C c F̃c,N
Q̃s,N
Qs,N = 1 P (3)
S s Q̃s,N
1
A detailed description of the calculation of value-added exports on sectoral level is in Koopman et al.
(2014).
2
A list of included countries is provided in the appendix. The three countries for which data is not
available in every complexity metric, i.e. LUX, MLT and TWN, are excluded from the empirical
assessment. P
3 EXPcp / EXPcp
Specifically, country c has a Revealed Comparative Advantage in product p, if P EXPcp / Pp EXPcp >
c c,p
1.
4
Tacchella et al. (2012) suggest this weighting scheme as a robustness check for the binary adjacency
matrix, and refers to the nominator of the Revealed Comparative Advantage (Balassa, 1965).
5
However,
P adapting the ECI accordingly is not possible, since, by definition of the weighting matrix,
c Wcs = 1 ∀c. This restricts any variation across industries in the calculation of ECI as ks,N = 1
∀N (see Hidalgo & Hausmann, 2009). Moreover, applying the ECI to a binary industry-country
adjacency matrix yields results that highly depend on the choice of N .
3F̃c,N and Q̃s,N are defined as
X
F̃c,N = Wcs Qs,N −1 (4)
s
1
Q̃s,N = P (5)
c Wcs (1/V XFc,N −1 )
The starting values are set to Q̃s,0 = 1 ∀s and F̃c,0 = 1 ∀c. F̃c,N describes the weighted
sum of industry complexity levels Qs,N , weighted by the share of value-added exports
country c has in the respective industry. After normalizing F̃c,N at every iteration,
V XFc,N denotes the complexity level associated with country c. The auxiliary variable
Q̃s,N shows that an industry’s complexity is positively related to the complexity of
countries exporting significant value-added in that industry. Due to the normalization
at every step, V XFc,N and Qs,N converge to a unique value for every c or s, respectively.
3 Empirical assessment
In this section, I compare the proposed complexity metric VXF to the two established
indices based on gross exports. As Table 1 shows for 2014, the complexity country
rankings differ substantially. While the United States top the rankings if complexity is
approximated by value-added exports, EF and ECI find China and Japan to be the most
complex country, respectively, in 2014. Canada and the Netherlands are, for example,
among the top 10 countries in terms of VXF, but are not present among the top 10 of
the other metrics. For China, the high complexity is driven by a high share of value-
added exports in complex manufacturing industries such as electronics, but also by the
definition of China in the data, which includes Hong Kong and Macao. This increases
the diversity, and thus complexity, considerably due to a high share of value-added in
financial services and R&D.
The indices seem to vary in their explanatory power of GDP per capita growth rates.
For a preliminary inspection, Figure 1 displays the unconditional correlation between
the growth of each of the complexity metrics and GDP per capita growth between 2000
and 2014. It can be seen that the empirical link between VXF and economic growth is
stronger with an R2 of 0.63 compared to 0.31 (ECI) or 0.24 (EF ).
For further investigation, I create a panel dataset including 40 lower-middle- to high-
income countries and comprising three non-overlapping time periods: 2000-2004, 2005-
2009, and 2010-2014. I estimate the following first-differenced growth model accounting
4Rank VXF EF ECI
1 USA CHN JPN
2 CHN DEU DEU
3 DEU JPN CHE
4 GBR ITA SGP
5 JPN USA KOR
6 FRA FRA AUT
7 KOR ESP SWE
8 ITA IND CZE
9 CAN BEL FIN
10 NLD GBR HUN
Table 1: The ten most complex countries for VXF, EF and ECI in 2014
(a) VXF (b) EF (c) ECI
Figure 1: Unconditional correlation between complexity and GDP per capita growth
for individual and time fixed effects
yi,t − yi,t−4 = αyi,t−4 + β1 (Ci,t − Ci,t−4 ) + β2 Ci,t−4 + γXi,t + θt + (i,t − i,t−4 ) , (6)
where C denotes one of the three complexity metrics and y the logarithmic GDP per
capita. Xi,t includes typical growth model covariates, i.e. population growth (n), change
in capital (K), change in human capital (H), and the initial stock of human capital.
Columns (1)-(3) of Table 2 show the regression results. It shows that, on the one hand,
the conditional correlation is stronger for V XF than for metrics based on gross exports.
On the other hand, including V XF instead of EF or ECI heightens the explanatory
power of the model considerably. Moreover, the lower point estimate for the initial
human capital stock, if V XF is included, may allow for the conclusion that complexity
in terms of a country’s value-added export structure more accurately depicts capabilities.
Keeping in mind the caveats attached to dynamic panel models, I estimate a non-
5dynamic fixed effects model as a robustness check.6 Columns (4)-(6) in Table 2 show the
results. While the impact of complexity in terms of gross exports on growth becomes
insignificant in this specification, both for EF and ECI, V XF is still significantly
correlated with economic growth.
Furthermore, the approximation of economic complexity based on value-added exports
may be a source of endogeneity, since they directly represent a part of GDP. However, the
iterative processes described in Equations 2 to 5 quantify the structure of an economy’s
value-added exports. That is, an increase in value-added exports does not automatically
lead to an increase in V XF . Rather, this depends on whether it positively affects the
complexity of the value-added export basket, either by an increase in its diversification
or by a further specialization in a complex industry (see Equation 1). To further support
that the relationship between V XF and GDP per capita growth is not driven by simply
exporting more and thereby increasing value-added, I additionally control for a country’s
trade openness.7 Table A.1 in the appendix shows that the results remain qualitatively
unchanged, if holding trade openness constant.
4 Conclusion
This article contributes to the literature by providing a new perspective on complexity.
Established metrics approximate an economy’s complexity by its gross export structure.
I suggest using a country’s value-added export structure instead, since value-added ex-
ports are arguably a more reliable depiction of economic performance. Based on a
weighted adjacency matrix and iterative processes analogous to Tacchella et al. (2012)
and Tacchella et al. (2013), I introduce the value-added export Fitness (V XF ) metric.
I show that V XF , firstly, leads to substantially different complexity rankings compared
to the established ECI and EF indices. That is, the United States top the rankings
in terms of V XF , while Japan and China are the most complex countries in terms of
ECI and EF , respectively. Secondly, including V XF , instead of ECI or EF , in a
first-differenced growth model with fixed effects controlling for typical growth regression
covariates, improves the explanatory power considerably.
Further research may rely on different inter-country Input-Output Tables to investi-
gate a larger number of countries, including, specifically, more low- and middle-income
6
A Generalized Method of Moments (GMM) approach may be preferred, but System-GMM estimations
lead to overidentification issues due to the small number of countries.
7
The applied measure for trade openness is the KOF de-facto trade openness index, which takes exports
and imports of goods and services as well as trade partner diversification into account (Gygli, Haelg,
Potrafke, & Sturm, 2019).
6Dependent variable: yi,t − yi,t−4
(1) (2) (3) (4) (5) (6)
∗∗∗ ∗∗∗ ∗∗∗
yi,t−4 −0.254 −0.108 −0.120
(0.062) (0.038) (0.044)
ni,t −0.044 0.476 0.179 −0.381 −0.193 −0.307
(0.283) (0.395) (0.366) (0.298) (0.357) (0.344)
∆log(K)i,t 0.189∗∗∗ 0.233∗∗∗ 0.235∗∗∗ 0.204∗∗∗ 0.241∗∗∗ 0.245∗∗∗
(0.021) (0.022) (0.025) (0.022) (0.021) (0.022)
∆log(V XF )i,t 0.270∗∗∗ 0.137∗∗∗
(0.044) (0.032)
log(V XF )i,t−4 0.159∗∗∗
(0.030)
∆log(EF )i,t 0.112∗ 0.073
(0.068) (0.067)
log(EF )i,t−4 0.069∗∗
(0.030)
∆ECIi,t 0.090∗∗ 0.045
(0.036) (0.033)
ECIi,t−4 0.064∗∗
(0.031)
∆log(H)i,t −0.099 −0.198 −0.045 −0.399 −0.717∗∗ −0.644∗∗
(0.304) (0.339) (0.312) (0.293) (0.301) (0.287)
log(H)i,t−4 0.619∗∗∗ 0.772∗∗∗ 0.736∗∗∗
(0.209) (0.218) (0.217)
Observations 120 120 120 120 120 120
R2 0.690 0.585 0.577 0.577 0.500 0.496
Adjusted R2 0.480 0.304 0.291 0.319 0.195 0.190
Note: All regressions include individual and time fixed effects. Standard errors accounting
for heteroskedasticity are applied. ∗ pcountries. A larger sample may allow for a more robust econometric assessment of the
effects of complexity on economic growth. Other currently available databases, however,
use a coarser industry classification. Moreover, future research should more thoroughly
investigate the interactions between human capital and complexity.
Acknowledgements
I am thankful to two anonymous referees, as well as to Jesús Crespo Cuaresma, Wolfgang
Schwarzbauer, Johannes Berger, Ludwig Strohner, Martin Wolf, Michael Berlemann, and
Tobias Thomas for helpful comments.
Appendix
List of countries (ISO 3166-1). AUS, AUT, BEL, BGR, BRA, CAN, CHE, CHN,
CYP, CZE, DEU, DNK, ESP, EST, FIN, FRA, GBR, GRC, HRV, HUN, IDN, IND,
IRL, ITA, JPN, KOR, LTU, LVA, MEX, NLD, NOR, POL, PRT, ROU, RUS, SVK,
SVN, SWE, TUR, USA.
References
Balassa, B. (1965). Trade liberalisation and revealed comparative advantage. The
Manchester School , 33 (2), 99–123. doi: 10.1111/j.1467-9957.1965.tb00050.x
Feenstra, R. C., Inklaar, R., & Timmer, M. P. (2015). The next generation of the
penn world table. American Economic Review , 105 (10), 3150–3182. doi: 10.1257/
aer.20130954
Gygli, S., Haelg, F., Potrafke, N., & Sturm, J.-E. (2019). The kof globalisation index–
revisited. The Review of International Organizations, 14 (3), 543–574.
Hausmann, R., Hwang, J., & Rodrik, D. (2007). What you export matters. Journal of
Economic Growth, 12 (1), 1–25. doi: 10.1007/s10887-006-9009-4
Hidalgo, C. A., & Hausmann, R. (2009). The building blocks of economic complexity.
PNAS , 106 (26), 10570–10575.
Hidalgo, C. A., Klinger, B., Barabási, A.-L., & Hausmann, R. (2007). The product space
conditions the development of nations. Science (New York, N.Y.), 317 (5837), 482–
487. doi: 10.1126/science.1144581
8Dependent variable: yi,t − yi,t−4
(1) (2) (3) (4) (5) (6)
∗∗∗ ∗∗∗ ∗∗∗
yi,t−4 −0.278 −0.150 −0.185
(0.057) (0.037) (0.037)
ni,t −0.001 0.487 0.168 −0.387 −0.265 −0.384
(0.271) (0.377) (0.356) (0.290) (0.354) (0.357)
∆log(K)i,t 0.178∗∗∗ 0.216∗∗∗ 0.213∗∗∗ 0.200∗∗∗ 0.238∗∗∗ 0.237∗∗∗
(0.021) (0.021) (0.022) (0.022) (0.020) (0.021)
∆log(V XF )i,t 0.264∗∗∗ 0.128∗∗∗
(0.043) (0.030)
log(V XF )i,t−4 0.162∗∗∗
(0.028)
∆log(EF )i,t 0.081 0.044
(0.063) (0.062)
log(EF )i,t−4 0.065∗∗
(0.026)
∆ECIi,t 0.118∗∗∗ 0.056
(0.037) (0.035)
ECIi,t−4 0.087∗∗∗
(0.028)
∆log(H)i,t −0.226 −0.322 −0.242 −0.509∗ −0.825∗∗∗ −0.740∗∗
(0.305) (0.343) (0.312) (0.289) (0.308) (0.302)
log(H)i,t−4 0.581∗∗∗ 0.746∗∗∗ 0.662∗∗∗
(0.183) (0.181) (0.176)
∆log(KOF )i,t −0.081∗∗ −0.135∗∗∗ −0.152∗∗∗ −0.096∗∗∗ −0.116∗∗∗ −0.120∗∗∗
(0.035) (0.035) (0.036) (0.036) (0.035) (0.036)
Observations 117 117 117 117 117 117
R2 0.718 0.628 0.641 0.608 0.537 0.543
Adjusted R2 0.518 0.366 0.388 0.359 0.243 0.253
Note In contrast to Table 2, these regressions do not include ROU due to availability of the
trade openness indicator. All regressions include individual and time fixed effects. Standard
errors accounting for heteroskedasticity are applied. ∗ pJohnson, R. C., & Noguera, G. (2017). A portrait of trade in value-added over four
decades. The Review of Economics and Statistics, 99 (5), 896–911. doi: 10.1162/
REST a 00665
Koopman, R., Wang, Z., & Wei, S.-J. (2014). Tracing value-added and double counting
in gross exports. American Economic Review , 104 (2), 459–494. doi: 10.1257/
aer.104.2.459
Tacchella, A., Cristelli, M., Caldarelli, G., Gabrielli, A., & Pietronero, L. (2012). A new
metrics for countries’ fitness and products’ complexity. Scientific reports, 2 , 723.
doi: 10.1038/srep00723
Tacchella, A., Cristelli, M., Caldarelli, G., Gabrielli, A., & Pietronero, L. (2013). Eco-
nomic complexity: Conceptual grounding of a new metrics for global competi-
tiveness. Journal of Economic Dynamics and Control , 37 (8), 1683–1691. doi:
10.1016/j.jedc.2013.04.006
The Growth Lab at Harvard University. (2019). Growth projections and complexity
rankings. doi: 10.7910/DVN/XTAQMC
Timmer, M. P., Dietzenbacher, E., Los, B., Stehrer, R., & de Vries, G. J. (2015).
An illustrated user guide to the world input-output database: The case of global
automotive production. Review of International Economics, 23 (3), 575–605. doi:
10.1111/roie.12178
Timmer, M. P., Miroudot, S., & de Vries, G. J. (2019). Functional specialisation in
trade. Journal of Economic Geography, 19 (1), 1–30. doi: 10.1093/jeg/lby056
Wittgenstein Centre for Demography and Global Human Capital. (2018). Wittgen-
stein centre data explorer version 2.0 (beta). Retrieved from http://www
.wittgensteincentre.org/dataexplorer
10www.ecoaustria.ac.at
You can also read
